04-01-2013, 04:01 PM
Content-Based Image Retrieval System Based on Self Organizing Map, Fuzzy Color Histogram and Subtractive Fuzzy Clustering
1Content-Based Image.pdf (Size: 528.79 KB / Downloads: 29)
Abstract:
A novel system with high level of retrieval accuracy has been presented in this paper. Color as one of the most
important discriminators in CBIR (content-based image retrieval) is utilized through calculating some of the primitive color
features. The indexing of image database is performed with SOM (self-organizing map) which identified the BMU's (best
matching units). Subsequently, Fuzzy Color Histogram (FCH) and subtractive fuzzy clustering algorithms have been utilized to
identify the cluster for which the query image is belonging. Furthermore, the paper presents an enhanced edge detection
algorithm to remove unwanted pixels and to solidify objects within images which ease similarity measures based on extracted
shape features. The proposed approach overcomes the computational complexity of applying bin-to-bin comparison as a multi
dimensional feature vectors in the original color histogram approach and improves the retrieval accuracy based on shape as
compared with the most dominant approaches in this filed of study.
Introduction
In Content-Based Image Retrieval (CBIR), researchers
seek for efficient and robust methods to retrieve
relevant images from huge images database utilizing
automatic derivation of local and global features from
image query as well as images database. Features as
shape, color, and texture are the most dominant
features to be considered. There are many similarity or
dissimilarity measures to rank the retrieved images
based on its relevancy to the query image.
Previous Work
In [2], they propose probabilistic framework to process
multiple image queries. The proposed framework is
independent from similarity measures and gives rise to
a relevance feedback mechanism. In [26], CBIR
method to diagnose aid in medical images is proposed.
Images are indexed without extracting domain-specific
features; a signature is built for each image via wavelet
transform. In [10], they propose two CBIR frameworks
based on genetic programming. The first framework is
concerned with user indication of relevant images,
while the second one considers the relevant and nonrelevant
indicated images. In [27], new multiresolution
fusion Algorithm for spatially registered
multi-sensor fusion is proposed.
Paper Outline
In this paper, a new approach for color and shapebased
image retrieval based on SOM and subtractive
fuzzy clustering Algorithm is presented. The rest of
the paper is organized as follows. Section 2 is an
overview of the proposed approach. Shape features
extraction and Algorithms are presented in section 3.
Section 4 illustrates similarity measures and
performance evaluation. Section 5 presents the
experimental results. Finally, the conclusion is drawn
in section 6.
Image Segmentation and Fuzzy Color
Histogram
Traditional color histogram approach does not take
into consideration the color similarity across different
bins (shades) or the color dissimilarity in the same bin.
Fuzzy color histogram approach [31] has many
advantages over the conventional color histogram
approach. Fuzzy Color Histogram (FCH) considers the
color similarity of each pixel's color associated to all
the histogram bins through fuzzy set membership
function such as the degree of "belongingness". As
compared to the traditional color histogram approach
which assigns each pixel into one of the bins only
Edge Detection Enhancement
Many of edge detectors are available to researchers
[19]. Marr and Hildreth convolve a mask over the
image and label zero-crossings of the convolution
output as edge points [20]. In [12], an approach
combining contrast threshold and analysis of direction
dispersion to find edges is presented. In [3], they label
peaks in the magnitude of the first derivative of the
intensity profile along a scan-line as feature points for
matching. Other popular gradient edge detectors are
the Canny, Roberts, Sobel and Prewitt operators [4].
Comparing objects based on edge operators only does
not yields to satisfactory results in most cases. That
because if there is any variation in image brightness,
then the same image looks different after applying the
edge operator. Moreover, the unwanted pixels in the
image affect the retrieval accuracy dramatically. In this
research and in order to overcome some of these
problems an Algorithm to filter the images at the preprocessing
stage is proposed.
Experimental Results
The proposed approach is tested on a general purpose
image database with 1000 images from COREL. The
1000 images are classified to 10 categories with 100
images each. Five images are randomly selected from
each category (e.g., Dinosaur, Beach, and Vehicles). A
retrieved image represents a correct match if and only
if it belongs to the same category as the query image.
The average precision is calculated through evaluating
the top 20 returned results. Due to space limitations,
only the top 9 matches to query images are shown in
Figure 3.
Conclusions
Region based segmentation and image clustering
combined with edge detection enhancement is
promising approach in CBIR. In this research the
integration of different methods in CBIR succeeds to
achieve robust, reliable, and a high level of retrieval
accuracy system. The experimental results on 1000
images from COREL database show that the proposed
approach achieves high retrieval accuracy with
valuable reduction to the number of extracted features.
Moreover, the comparisons with traditional color
histogram, ECR, and UFM methods prove that the
proposed approach is able to improve the accuracy of
retrieval dramatically.